Influence of Coerced Input on English Learners’ Grammar

Influence of Coerced Input on

English Learners’ Grammar*1)

Soyeon Yoon

(Incheon National University)

Yoon, Soyeon. (2016). Influence of Coerced Input on English Learners’ Grammar. Language Research, 52.1, 57-85.

This study examines if speakers’ existing grammar can change after the speakers experience the input that does not match their linguistic knowledge, i.e., the coerced input, on the assumption of the us-age-based model. Specifically, sentences that are composed of a di-transitive construction and main verbs requiring various degrees of co-ercion were created. Korean speakers judged the acceptability of these coerced sentences (Pretest), read the sentences appearing in the pas-sages eight times over four weeks (input sessions), and judged the ac-ceptability of the coerced sentences once again (Posttest). The results are as follows: In the Posttest they judged the coerced sentence that they read in the input sessions more acceptable. Moreover, in the Posttest they generalized the pattern that they received from the input: When they read the sentences that are composed of verbs and NPs dif-ferent from the input sentences, they judged the sentences more acceptable. The results suggest that recurrent experience of coerced in-put can change existing grammar of the English learners.

Keywords: coercion, acceptability judgment, usage-based model, grammar change

1. Introduction

In Construction Grammar, coercion is defined as the resolution of se-

mantic incompatibility between a lexical item and a construction in which

the lexical item occurs (Croft 1991; Michaelis 2005; Pin͂ango et al. 2006;

Ziegeler 2007). For example, the verb give is semantically compatible with

* This work was supported by the Incheon National University Research Grand in 2014. I would like to thank three anonymous reviewers of this journal for their helpful and insightful comments.

58 Soyeon Yoon

the ditransitive construction (DC) [SUBJi [V NP1j NP2k]] because the

meaning of give is almost identical to the meaning of the construction

in that both indicate ‘transfer of possession of k from i to j. On the other

hand, cut is not compatible with the DC because cut typically involves

only two arguments (Agent and Patient) and does not involve the meaning

of transfer. However, when it is used in the sentence I cut him a belt

(Yoon 2012; 2013), cut is coerced to mean ‘to make something by cutting

and transfer it to someone,’ and the incompatibility is resolved.

The degree of semantic compatibility between a verb and a construction

and their coercion are closely related with frequency of usage (Yoon 2012;

2013). Specifically, if a verb is semantically more compatible with a partic-

ular construction, their co-occurrence involves less coercion, and they

are used more frequently used together. However, this correlation only

shows that the grammatical knowledge and language use are correlated.

Thus, bi-directional interpretation is possible: A particular verb and a

construction are recognized compatible because speakers experience the

pattern frequently; the verb and the construction are used together fre-

quently because they are compatible. In other words, the correlation does

not indicate whether the frequent usage affects the linguistic knowledge

of semantic compatibility.

Regarding the influence of the usage on the grammar, the usage-based

model, proposed by Langacker (1988), assumes that linguistic knowledge

or grammar is grounded in language use. The model predicts that if ex-

pressions of a similar pattern are used together frequently, the pattern

is cognitively entrenched as parts of grammar. Following this prediction,

we can test the hypothesis that language use affects grammar. Specifically,

if speakers continuously experience instances where a verb and a con-

struction are not very compatible, this experience affects the acceptability

judgments on the instance where the incompatible verb and construction

co-occur. If empirical evidence supports this hypothesis, it strongly sug-

gests that frequency of language use does affect grammar. The current

study attempts to show that language use can extend and change grammar.

To this end, it investigates to what extent English learners can accept

coerced expressions and whether the inputs that are slightly different from

Influence of Coerced Input on English Learners’ Grammar 59

the existing grammar, i.e. coercion, can be generalized as a part of grammar

after they recurrently experience coerced expressions.

2. Research Background

2.1. Usage-based Model

In Cognitive Grammar (Langacker 1987; 1988), on which the us-

age-based model is built, a “schema” is a generalized pattern that speakers

extract from similar repeated instances. It is not a mechanism that produces

“outputs” as the “rules” do in Generative Grammar. Rather, it serves

as symbolic resources which speakers exploit to construct new expressions

(Langacker 1988: 132). Schemas function as a device for categorization

and specify the properties that the instances have to have in order for

them to be valid members of the category. For example, the DC specifies

a verb which has the meaning of transfer of possession. In turn, the verb

give specifies a construction which means transfer of possession, i. e.,

the DC. We can determine whether or not to combine two linguistic

components based on how well the instance where the components occur

together is categorized by the schemas of those components. For example,

the instance where give is used in the DC, such as I gave John a book,

is categorized as a member of the schema of the DC and the schema

of give because the instance fits both the constructional schema and the

lexical schema. Therefore, we can conclude that give and the DC can

co-occur with no problem.

There are two kinds of categorizing relationships. One is “elaboration”

or “instantiation” of a schema. A schema is elaborated by an instance

that conforms to or is compatible with the specifications of the schema.

The example of elaboration is I gave John a book: It is an instantiation

of the give schema and the DC schema.

The other kind of categorizing relationship is “extension.” In this case,

an instance is not perfectly compatible with the schemas but they can

be roughly viewed as a member of the same category if we extend the

schema. For example, the DC typically specifies a verb of transfer of pos-

60 Soyeon Yoon

session, but cut is not an instantiation of these verbs. However, it can

be viewed as one of the verbs that can occur in the DC when its meaning

is “accommodated.” Accommodation is Langacker (1988)’s terminology,

which means the adjustment of a component’s details when integrated

with another incompatible linguistic component. The concept of coercion

used in Construction Grammar can be seen as one type of accommodation

process. When the meaning of cut is accommodated or coerced to mean

‘to make something by cutting and transfer it to someone,’ it can be used

as an extension of the verb schema that occur with the DC.

The elaboration and extension of the schema are related with frequency

of language use: If speakers experience similar instances repeatedly, the

pattern is entrenched as a schema. The verb give or the verbs in the similar

class (e.g., hand and lend) and the DC are used very frequently, and thus

their co-occurrence is entrenched and recognized as compatible. On the

other hand, cut and the DC are not used frequently, and these two are

recognized as much less compatible. In short, in the usage-based model,

the frequency of language use is important in forming grammar.

Following the assumption of the usage-based model we can make a

prediction as follows: If speakers experience the expression where the

DC and a less compatible verb such as the cutting verb (e.g., cut) are

used together, they have to extend the schemas of the DC and the verb

or coerce their co-occurrence; they repeatedly coerce the similar instances

by accommodating the schemas of this less compatible verb and the DC,

the extended schemas can be entrenched. Therefore, the cutting verbs

and the DC are recognized more compatible. In other words, the grammar

may change in accordance with the frequent input which is modified

from the existing grammar.

2.2. Usage-based Model and Language Acquisition

In usage-based approaches to language acquisition (Kidd, Lieven, and

Tomasello 2010; Boyd and Goldberg 2009; Year and Gordon 2009; Ellis,

O’Donnell, and Römer 2014), grammar is acquired through gen-

eral-cognitive learning principle, such as categorization and generalization


of the instances, and salience and prototypicality of the input. Thus, the

importance of input has been emphasized, and the frequency and the

types of the input have been extensively discussed. For example, Goldberg,

Casenhiser, and Sethurman (2004) investigated three constructions -- the

intransitive motion construction (e.g., The fly buzzed into the room), the

caused-motion construction (e.g., Pat sneezed the foam off the cappuccino),

and the DC, and showed that most of the input that children received

from their caregivers was skewed towards several general verbs that are

used in each construction (e.g. go, put, and give, respectively). Kidd, Lieven,

and Tomasello (2010) also supported exemplar-based learning, showing

that children’s experimental performance was better when the sentence

contained high frequency verbs. These studies show that the verb that

is the most frequently used in the construction is representative of other

verbs in the similar verb categories, and children acquire the construction

based on the most frequent verb. As opposed to these studies, by using

a novel construction [V N1 N2] whose novel meaning is “N2 approaches

N1,” Wonnacott, Boyd, Thomson, and Goldberg (2012) showed that chil-

dren could generalize the pattern to unfamiliar verbs if the input contains

various verbs used in the construction.

Regarding the acquisition of the verb-construction in the second lan-

guage acquisition, Ellis and Ferreira-Junior (2009) claimed that more typi-

cal verb exemplars will contribute to defining the verb categories that

the verb belongs to, and this exemplar will be recognized as a prototype

of the verb category (Ellis and Collins 2009). Ellis, O’Donnell, and Römer

(2014) addressed that prototypical inputs are more advantageous in the

verb-argument structure because the prototype occurs the most frequently

in usage, and it gets the network centrality, which makes the associated

verbs linked with the construction. These studies show that skewed input

plays important roles in the verb-construction acquisition. Year and

Gordon (2009), however, presented that a more balanced set of verbs

facilitated the verb-construction acquisition, which is in line with the find-

ings of Wonnacott, Boyd, Thomson, and Goldberg (2012). In their experi-

ments on Korean children learning English, the children could use the

DC with a variety of verbs for a longer time when the inputs were a

62 Soyeon Yoon

balanced set of verbs, compared to when the inputs were skewed towards

prototype (e.g. give in the DC).

These aforementioned studies on the verb-construction acquisition

mainly examine the roles of the inputs in the situation where the speakers

initially establish the grammar of the verb-construction. They mostly deal

with grammatical inputs (i.e., semantically compatible verbs and con-

structions), and showed how the grammatical inputs are used (e.g., skewed

inputs or balanced inputs) and what roles they play in the first and second

language acquisition. However, they do not address how existing grammar

can change depending on the inputs. If speakers frequently encounter

relatively incompatible verbs-construction co-occurrences, and if these in-

puts affect speakers’ judgments about compatibility, this suggests that the

frequent experience of linguistic expressions that do not match speakers’

existing grammar does affect their linguistic knowledge and eventually

changes grammar. Showing that recurrent experience of the coerced input

changes grammar will give some light not only to the studies on language

change but also to the studies on the influence of input on L2 learners.

3. Experiment Design

In the experiment, Korean speakers were asked to judge the acceptability

of English sentences where a verb was used in the DC ([V NP1 NP2])

as a main verb. The selected verbs varied with regard to the semantic

compatibility with the DC. The experiment tested if the sentences of vari-

ous degrees of compatibility are judged more acceptable when speakers

experience these sentences repeatedly, and this increased acceptability in

the judgments can be generalized to the cases where the DC is used with

other NPs or other verbs in the syntactically / semantically similar verb

classes. The prediction is as follows: The input sentences will be judged

more acceptable after several times of exposure to the input, and also,

the sentences where different NPs or other verbs were used will be judged

more acceptable.

The experiment was divided into three parts ― preliminary test with

native speakers of English (NS), Pretest of acceptability judgments with


non-native speakers (NNS), input sessions where the sentences with differ-

ent degrees of compatibility were exposed to the NNS, and Posttest of

the acceptability judgments with NNS.

3.1. Participants

Fifty eight university students participated in the experiment. They were

native speakers of Korean, and have learned English for more than ten

years. It is anticipated that NNS can show the influence of the coerced

inputs on linguistic knowledge. They have not experienced coercion as

much as NS do because their English learning has been focused on acquir-

ing prescriptive grammar while we can find the cases of coercion not

infrequently in the NS corpora (Yoon 2012; 2013). Therefore, it is expected

that the NNS would hesitate to accept the co-occurrences of less compat-

ible elements and judge the coerced sentences less acceptable than the

NS do. If the NNS judge the coerced sentences more acceptable after

they are repeatedly exposed to the coerced input despite their linguistic

knowledge, the results will strongly suggest that the frequent input does

change grammar.

The participants were divided into two Groups, each of which had

twenty nine people: Target group which was exposed to the sentences

where the DC was used with the main verb of various degrees of compati-

bility, and Control group which was exposed to the sentences irrelevant

to the DC. The students were randomly assigned to one of the two groups.1)

Additionally, in order to see if the Korean speakers’ coercion is different

from the native speakers of English, seven NS participated in the prelimi-

nary test. Since the number of NS was not enough for any inferential

1) In the current experiment, the participants’ English proficiency was not considered. Prior to this experiment, the same experiment had been conducted considering the students’ English proficiency. There were 19 people in each level (i.e. high vs. low) and group (Target and Control). The participating students were different from the current participants. In this previous experiment, the participants received the input four times, which is different from the current experiment in which the participants received input eight times. The result was that there was no significant difference be-tween the groups and between the levels. Thus, in the current experiment, I did not divide the level of proficiency, but instead, I increased the number of participants in each group and the number of input.

64 Soyeon Yoon

statistics, I simply compared the means resulting from the preliminary

test of the NS with the means resulting from the Pretest of the NNS.

3.2. Stimuli

The verbs that were to be used in the main verb slot in the DC were

selected from the five categories that are divided depending on the degree

of semantic compatibility with the DC (Yoon 2012; 2013). Among the

verb categories, the verbs of inherent transfer (e.g., give, send, lend) were

the most compatible with the DC, and the verbs of events internal to

the Agent (e.g., remember, imagine, stay) were the least compatible. The

verbs of possible transfer (e.g., make, cook, find), verbs of prevented transfer

(e.g., refuse, deny, decline), and verbs of impossible transfer (e.g., cut, slice,

break) were in the middle, respectively. Five verbs were selected, and each

of them represented each of the five categories: give, make, refuse, cut,

and remember, respectively.

This verb selection is supported by the frequency of their usage in the

DC. According to Yoon (2013), the degree of semantic compatibility was

correlated with the frequent usage in the corpus. For example, when the

49 verbs used in the DC in the part of British National Corpus were

ranked in the order of frequent usage in the DC, give was the first, make

was the fourteenth, and refuse was the thirtieth, while cut and remember

were not used in the DC.

In the experiment, these five verbs were selected and five sentences

were created as in (1).

(1) a. Jenny gave me a rose at the restaurant.

b. Bill made her a toy for her birthday.

c. Sam refused her a ring last night.

d. Jane cut him a belt on his birthday.

e. Beth remembered me a watch on their anniversary.

I asked some of the NS participants to paraphrase the sentences in (1).

In general, the coerced interpretation was that the subject gave NP2 to


NP1 in the manner of the action designated by the verb. Specific inter-

pretations are given in (2).

(2) a. ----

b. Bill made a toy and gave it to her.

c. Sam refused to give her a ring.

d. Jane cut a belt out of a piece of leather and gave it to him.

e. ----

Note that (1a) does not require a coerced interpretation, and (1e) cannot

be coerced at all (i.e., the NS could not comprehend (1e) at all, and

judged it ‘not acceptable at all’ in the preliminary test, which will be

discussed in 4.1).

Of the five sentences, (1b) - (1e) were continuously given to the Target

group participants in different sixteen passages eight times (following Year

and Gordon 2009) during the input sessions. It is controversial whether

skewed input or balanced input plays more important roles in generalizing

the acquired pattern as was discussed in 2.2. Nevertheless, as the first

step to investigate the influence of the coerced input, the input that is

used in the current study is skewed to one verb in each category.2) Note

that the sentence with give was excluded from the input sessions because

it is assumed to be very compatible with the DC, and thus, frequent

input of give was expected not to affect the acceptability judgments. The

subject and the NP1 (Recipient) varied depending on the content of the

passage, but the verb and the corresponding NP2 used in (1b) - (1e) were

the same throughout the sessions.

To examine if participants can generalize the pattern that they received

through the input sessions, in the Pretest and the Posttest, I tested two

more sentences for each verb in (1) in which the NP2 varied; one takes

the NP2 of edible entity (e.g., an onion, a cup of tea) and the other any

non-edible entity (e.g., a table, a knife). In addition, I selected two more

verbs for the verb in each degree of semantic compatibility. When selecting

2) Further study should be followed regarding the influence of coercion when the input is balanced.

66 Soyeon Yoon

these verbs, I referred to the verb classes which were categorized by Levin

(1993) according to their similar semantic and syntactic properties. To

see the generalization of the input pattern is further extended, two more

verbs were added to each of the five verbs in (1); one from the same

verb class, and the other from the similar verb class. In Table 1, the

verbs used in the stimuli sentences are summarized. The labels in the

parentheses follow those of Levin’s (1993).

Table 1. The Verbs Used in the Stimuli Sentences

Compatibility Categories

input verb x 3(NP2 used in the input / edible /

non-edible)

Class1(verbs in the same class)

Class2(verbs in the similar class)

Verbs of inherent transfer

give (give verb) lend (give verb) send (send verb)

Verbs of possible transfer

make (build verb) cook (build verb) find (get verb)

Verbs of prevented transfer

refuse3) deny decline4)

Verbs of impossible

transfer

cut (verbs of cutting)

slice (verbs of cutting)

break (verbs of

separating)

Verbs of events internal to the

Agent

remember (characterize verb)

imagine (characterize verb)

stay (exist verb)

Using the verbs in Table 1, I constructed 25 sentences. A part of the

stimuli sentences are given in (1), and the entire stimuli are presented

in Appendix 1.

I also added 50 fillers which are not the DC (i.e., caused-motion con-

struction and intransitive construction). The verbs in the filler were also

different from Table 1.

3) The verb refuse and deny were not labeled in Levin (1993, p. 47), but both were catego-rized as “non-alternating double object only” under “Dative alternation” section.

4) In this study, I included decline assuming that it denotes ‘prevented transfer.’ However, this verb is polysemous, so in Levin (1993) it was labeled as “verbs of caliberatable changes of state.” Since the latter meaning is far from the meaning intended by this study, I do not use the label.


3.3. Procedure

First, the 25 stimuli sentences and 50 filler sentences were given to

seven native speakers of English. The sentences were randomly ordered.

They rated the acceptability of the sentences on seven-point Likert scale

― 1 as the most acceptable and 7 as the least acceptable.

Second, fifty eight NNS rated the acceptability the 25 stimuli sentences

and 50 filler sentences (Pretest).

Third was the input session. The NNS were divided into two groups:

Twenty nine participants were in the Target group and the other twenty

nine in the Control group. Four input sessions were conducted for four

weeks. In each session, the participants read two sets of passages once

a week at the same time of the week at the same place. The passages

were selected from Corpus of Contemporary American English, forums,

blogs, and book excerpts that were publicized on the website, or newly

created.5) Each passage contained 203 words in average (ranging from

116 words to 252 words). The passages were modified for the purpose

of the experiment. One set of the passages was composed of two passages.

Each set that the target group read contained the input sentences in (1b-e)

where the subjects and NP1s were modified depending on the context

but NP2 was equivalent to those of (1b-e). Since there were two sets

of passages in each session, the input sentences appeared twice for the

Target group. In total, each input sentence was exposed to the participants

eight times (1 input per set * 2 sets per session * 4 sessions). Additionally,

in each session, there were four filler sentences which were selected from

the Pretest, and they were continuously used throughout the sessions so

that the participants do not particularly pay attention to the input

sentences. On the other hand, the passages that Control group read did

not contain the input sentences in (1b-e). Instead, the sentences were

deleted or replaced by other sentences.

Fourth, a week after the input sessions, both the Target group and

the Control group judged acceptability of the sentences used in the Pretest.

The scores from this Posttest were compared with those from the Pretest.

5) See Appendix 2 for the sources.

68 Soyeon Yoon

3.4. Prediction

The prediction of the comparison of the preliminary test results obtained

from the NS and the Pretest results obtained from the NNS is that NNS

will not judge the coerced sentences as acceptable as NS do because they

have not experienced the sentences with incompatible elements and have

learned only prescriptive grammar. Therefore, their judgments would be rather

binary: More compatible verbs, such as give, lend, send, make, cook, and find,

would be judged quite acceptable while less compatible verbs, such as cut,

slice, break, remember, imagine, and stay, would be not acceptable at all.

The prediction in the difference in Target group and Control group

in Pretest and Posttest is that only the Target group will judge the sentences

used in the input sessions (input sentences) more acceptable in the Posttest

compared to the Pretest. If the sentences are composed of the input verb

and the NP2 of edible (edible NP2 sentences) or non-edible entity (non-edi-

ble NP2 sentences), only the Target group will judge them more acceptable

in the Posttest because I expect that the participants can generalize the

pattern of the input sentences. However, this increased acceptability will

not be as great as the case of input sentences because these are slightly

different from what they experienced in the input session. For the sentences

of Class1 and Class2, only the Target group will judge the sentences more

acceptable in the Posttest, but again, the difference will not be as great

as the case of input sentences. The Control group will not show difference

in the judgments in Pretest and Posttest.

4. Results

4.1. Acceptability of Coerced Sentences: NS vs. NNS

I compared the preliminary test result of the NS with the Pretest result

of the NNS (both Target and Control). Since the number of NS was

not enough for an inferential statistics, I simply compared mean scores

of the sentences obtained from NS and NNS. For descriptive statistics,

see Appendix 3. Figure 1 to Figure 5 summarizes the mean of the accept-


ability judgment scores with 1 as the most acceptable and 7 as the least

acceptable. The solid line and the dotted line represent the scores of the

NS and NNS, respectively.

Figure 1. the Mean of input sentences (NS preliminary vs. NNS Pretest).

Figure 2. the Mean of the edible NP2 sentences (NS preliminary vs. NNS

Pretest).

Figure 3. the Mean of the non-edible NP2 sentences (NS preliminary

vs. NNS Pretest).

70 Soyeon Yoon

Figure 4. the Mean of Class1 sentences (NS preliminary vs. NNS Pretest).

Figure 5. the Mean of Class2 sentences (NS preliminary vs. NNS Pretest).

In Figure 1, Figure 3, and Figure 5, the acceptability scores of the NS

gradually increased as the verbs become less incompatible with the DC,

which means that the verbs selected in this experiment represent NS’s

linguistic knowledge about the degree of semantic compatibility: If a verb

is less compatible with the construction which the verb occurs in, the

sentence where the verb and the construction co-occur is judged less

acceptable.

However, if the NP2 was an edible entity, the acceptability was judged

better than the sentences with non-edible NP2s, as make (make NP1 a

cup of tea) and cut (cut NP1 a slice of bread) in Figure 2 show. Also, slice

in Class1 in this experiment was used with the NP2 of an edible entity

(slice NP1 an onion), which was judged quite acceptable as in Figure 4.

The prediction was that the action of cutting or slicing would damage

an entity, and this entity could not be transferred to someone else. However,

when it is used with an edible entity, it is easily interpreted as an action

of creating something for the purpose of serving. For example, we “create”

a slice of bread or slice an onion by cutting it so we can serve it to somebody.

As a result of the preliminary test of the NS and the pretest of the NNS,


we can see that the semantic compatibility between a verb and the DC

can be different depending on the semantics of the NP2.

Unlike the prediction that the NNS would judge the coerced sentences

less acceptable than the NS, the results show that the scores of both groups

were relatively similar as can be seen in Figure 1 to Figure 5. Specifically,

the scores and the line trend of NS and NNS shown in Figure 1, Figure

2, Figure 3, and Figure 4 are quite similar, unlike the prediction that

their judgments will be rather binary. This result implies that NNS might

have a similar intuition about the compatibility between the verbs used

in the experiment and the DC, which will be discussed in 5.

Interestingly, however, the least compatible sentences were judged more

acceptable than the NS did. Except for the least compatible verb in Class1

(i.e., imagine) in Figure 4, the scores of the verbs that were expected to

be the least compatible with the DC (i.e., remember and stay) diverged.

In other words, the NNS did not entirely reject the incompatible sentences.

The prediction was that the NNS would judge incompatible sentences

unacceptable. However, the result was that the NNS were careful to judge

incompatible verbs entirely unacceptable. It seems that the NNS were

not very confident in their linguistic knowledge about the compatibility,

and they might have thought that the NS could allow relatively incompat-

ible items to co-occur.

4.2. Comparison of the Pretest and Posttest (Target vs. Control)

The results generally confirmed the prediction that the Target group

would judge the input sentences more acceptable in the Postest, and this

increased acceptability would be applicable to the edible NP2, non-edible

NP2, Class1, and Class2 sentences.

For each sentence group (input, edible NP2, non-edible NP2, Class1,

and Class2), Repeated measure of ANOVA was used, in which Trial

(Pretest and Posttest) and Verb (five verbs from each compatibility cat-

egory) were within subject variables and the Group (Target and Control)

was a between-subject variable. The result is presented in Table 2.

72 Soyeon Yoon

Table 2. Mean Differences between the Pretest and Posttest for the input

Sentences (* for the p-value smaller than the significance level adjusted

for Bonferroni correction)

inputTarget Control

Pre Post t-test Pre Post t-test

give 2.03 1.41t(56) = 2.45,

p = .01*1.34 1.86

t(56) = -1.76, p = .04

make 2.96 2.27t(56) = 1.45,

p = .081.72 2.10

t(56) = -1.05, p = .15

refuse 4.79 4.00t(56) = 2.01,

p = .02*4.14 4.24

t(56) = -0.23, p = .41

cut 5.00 4.90t(56) =0.30,

p = .384.86 5.17

t(56) = -0.71, p = .24

remember 5.41 4.31t(56) = 3.27,

p =.00*5.17 4.69

t(56) = 1.18, p = .12

First, since the statistics in question was whether the Trial (Pretest and

Posttest) was different depending on the Group (Target or Control), only

the interaction of Trial and Group was examined. The result was that

only the input sentences showed significant interaction (F(1,56) = 6.03,

p < .05), whereas the other sentence groups did not. This means that

at least for the input sentences, the difference in Pretest scores and Posttest

scores was not equal across Target group and Control group. If we look

at the difference more closely regarding the input sentences, Pretest scores

and Posttest scores were different in Target group (M in Pre = 4.03, M

in Post = 3.38, F(1,28) = 8.36, p < .01) whereas those were not different

in Control group (M in Pre = 3.45, M in Post = 3.61, F(1,28) = .78,

p = .385). More specifically, the means of each verb in Pretest and Posttest

were compared through the pairwise t-test (one-tailed, because the assump-

tion was that the scores would be smaller (i.e., more acceptable) in the

Posttest). As a result, Target group showed increased acceptability. In

Table 2, the means of the sentences are presented. The significance level

was adjusted for sequential Bonferroni correction.6)

6) For each group (Target and Control), I conducted five t-tests. In order to decrease Type-I error, I adjusted the significance level following sequential Bonferroni correction. For the smallest p-value, I applied the significance level of .0100, and for


As we can see in Table 2, for the sentences where the main verbs were give, refuse, and remember, only the Target group judged the Posttest sentences more acceptable. Although not statistically significant, make was judged more acceptable. The Control group did not show significant difference. There were even cases where the score increased (give, make, refuse, and cut), meaning that the participants judged the sentences in the Posttest less acceptable.

This result suggests that when speakers experience the same expression recurrently, this expression is cognitively entrenched, and judged quite acceptable even though this expression does not perfectly agree with their existing linguistic knowledge.

If this recurrent experience is generalized as a pattern, we should be able to see that the sentences with different NP2s or Class1 verbs and Class2 verbs are judged more acceptable in the Posttest in Target group. Even though there was no effect of interaction of Trial and between-subject variable (Target / Control), when the Repeated Measure of ANOVA was conducted for Target and Control separately, the prediction was supported. Here, Trial (Pretest and Posttest) and Verbs (five verbs from each compati-bility category) were within subject variables. I examined if there was a main effect of Trial for Target and Control because the hypothesis was that there was significant difference between Pretest and Posttest. When there was a main effect of Trial, I further conducted one-tailed t-tests to compare the difference in Trial for each of the Verbs.

First, for the edible NP2 sentences, the Target group judged the Posttest sentences more acceptable (M in Pre = 3.41, M in Post = 3.03, F(1, 28) = 4.55, p < .05). Specifically, the sentences with remember was significantly different (M in Pre = 5.07, M in Post = 4.28, t(56) = 2.41, p < .01, significant even with Bonferroni correction). However, the Control group did not show difference (M in Pre = 3.60, M in Post = 3.46, the main effect of Trial, F(1,28) = .12, p = .73).

Second, for the non-edible NP2 sentences, both Target group (M in Pre = 4.12, M in Post = 3.43, the main effect of Trial, F(1,28) = 13.81, p < .01) and Control group (M in Pre = 3.39, M in Post = 2.92, the main effect of Trial, F(1,28) = 12.67, p < .01) showed main effect of

the next smallest one, .0125, .0167, .0250, and .0500, sequentially.

74 Soyeon Yoon

Trial. However, when examined closely with the t-test, the Target group’s judgment became more acceptable in the Posttest.

Table 3. Mean Differences between the Pretest and Posttest for the

Sentences with non-edible NP2 (* for the p-value smaller than the sig-

nificance level adjusted for Bonferroni correction)

non-edible NP2

Target Control

Pre Post t-test Pre Post t-test

give 2.07 1.55t(56) = 1.68,

p = .011.72 1.41

t(56) = 1.05, p = .15

make 2.79 2.21t(56) = 1.19

p = .123.14 1.86

t(56) = 2.65, p = .005*

refuse 5.55 4.93t(56) = 2.26,

p = .01*5.31 4.57

t(56) = 2.09, p = .02

cut 5.24 4.21t(56) =2.78, p = .003*

5.52 4.62t(56) = 2.46,

p = .01*

remember 4.97 4.24t(56) = 1.85,

p =.035.00 4.97

t(56) = .08, p = .47

In Table 3, when the significance level was .05, four verbs (give, refuse,

cut, and remember) were significantly different in Target and three (make,

refuse, and cut) in Control, even though two were significant in both groups

after Bonferroni correction.

Third, when the sentences were used with Class1 verbs, the Target

group judged the sentences in the Posttest more acceptable (M in Pre

= 3.73, M in Post = 3.40, the main effect of Trial, F(1,28) = 5.44, p

< .05) whereas there was no significant difference in Control group (M

in Pre = 3.43, M in Post = 3.55, the main effect of Trial, F(1,28) =

.45, p = .51). Specifically, when a verb slice was used in the DC, the

Target group’s score became smaller (M in Pre = 3.76, M in Post = 2.79,

t(56) = 2.00, p < .05), meaning that the Target group’s judgment became

more acceptable in the Posttest.

Finally, the sentences with Class2 verbs did not show any difference

either in the Target (M in Pre = 4.04, M in Post = 4.03, the main effect

of Trial, F(1,28) = .01, p = .92) or in the Control (M in Pre = 3.66,

M in Post = 3.79, the main effect of Trial, F(1,28) = .33, p = .57).


5. Discussion

Though we cannot statistically test the difference between the NS and

NNS due to the lack of the number of NS participants, the comparison

of the mean in the judgment scores in the preliminary test shows that

the NNS accept the coerced sentences similarly with the NS: Both NS

and NNS judged more compatible sentences more acceptable. I predicted

that the NNS in the experiment would show rather dichotomous judg-

ments (acceptable or not acceptable) because I assumed that most NNS

had been exposed to the coerced inputs less frequently than NS. Despite

the lack of exposure to the coerced inputs, the result of the NNS was

similar to that of the NS. One possible reason for this may be that NNS

were not confident in their grammatical knowledge in English, and thus

they tended to give more acceptable scores to each sentence than they

actually thought. For example they gave the score of 3 to one of the

coerced sentence when they actually thought it was 6. Interestingly, for

the least compatible sentences, the NNS seemed not very confident enough

with their linguistic knowledge about the compatibility to reject the

sentences. It means the possibility that the NNS may have more flexible

or less firmly established schemas of the verbs and the DC. If they experi-

ence coerced expressions more, they may readily accept them.7)

Another possible reason for the similar judgment pattern in NS and

NNS despite the different frequency of coerced input is that frequency

may not be the only crucial factor that affects English learners’ grammar.

As Yoon (2012, 2013) showed, acceptability judgment scores on the sen-

tences of different degrees of coercion are closely correlated not only with

the frequency of the verb-construction co-occurrence but also with the

semantic compatibility between the verb and the construction. As for the

similar judgment patterns, it is possible that the general knowledge regard-

ing the semantics was an important factor. As Wierzbicka (1996) claimed

universality in semantics, the semantic compatibility between English

verbs and constructions can be recognized by NNS as well. This may

7) This speculation could have been supported if there had been a follow-up interview with the NNS participants. Unfortunately, the experiment was not followed by an interview.

76 Soyeon Yoon

lead to the NNS’s similar judgment patterns with the NS despite the

lack of the coerced input. Note, however, that it does not indicate that

input is not important in language learning. As the results in the Posttest

shows, the input does affect grammar.

The results from the judgments of the NNS confirmed the prediction

that the speakers who recurrently experienced a particular coerced ex-

pression will accept the expression, and they can generalize a pattern

so that they can apply the generalization to similar expressions. On the

assumption of the usage-based model, the increased acceptability of the

coerced expression implies that speakers manage to extend the existing

schema in order to use incompatible items together.

As for the input sentences, the Target group judged relatively incompat-

ible sentences more acceptable after they are exposed to the input sentences

several times. This implies that the participants managed to extend their

existing schema of the verbs and the DC to use the incompatible elements

together, and they became familiar with these coerced expressions.

The acceptability judgments regarding the sentences other than input

also suggest the possibility that the speakers manage to make generalization

of the coercion. Even though the acceptability difference after the input

sessions was not as great as the case of the input sentences, when the

NPs varied, the participants seemed to judge the coerced sentences more

acceptable. It means that the speakers can extend the schema of the DC

and the input verbs relatively comfortably after they became more familiar

with the pattern. This generalization extends to the sentences where the

main verb is distinct from the input verbs. We could see that the partic-

ipants judged the Class1 sentences slightly more acceptable in the Posttest

although the difference was not as great as the case of the input sentences.

Unfortunately, we could not see any difference in Class2 verbs. Since

the Class2 verbs are semantically more distinct from the input verbs than

Class1 verbs, the Class2 sentences may require greater extension of the

existing schema. This may have led to no difference in acceptability

judgments. I expect that even Class2 verbs may be judged more acceptable

in the DC if speakers experience more instances of similar pattern of

the extension of the verb / construction schema.


It has been claimed that there is a correlation between language use

and linguistic knowledge that the verbs that are semantically more compat-

ible with a particular construction are used more frequently, processed

more rapidly, and judged more acceptable when they are used in the

construction (Yoon 2012; 2013). Expanding from this correlation, the

current study further claims that frequent usage of particular linguistic

elements (even if they are not perfectly compatible with one another)

affects the judgments about their co-occurrence, and ultimately implies

that language use changes linguistic knowledge about compatibility.

Coercion has been considered to represent one of the creative and dy-

namic aspects of language because it creates non-prototypical expressions

by combining incompatible linguistic items. If non-prototypical ex-

pressions are used frequently, grammar may change slowly over time.

For example, a beer has been discussed as a case of coercion because

a specifies an entity of [+bounded] while beer is [-bounded] (Michaelis,

2005). In other words, a and beer are incompatible. When they are used

together, however, we can coerce it to mean ‘beer in a particular container.’

When asked to NS, a beer is comfortably used in everyday life and does

not require extra processing (Ziegeler, 2007, Author, 2012). Similar exam-

ples are used pervasively as in a coke, a water, a wine, etc. It seems that

a can be used with an unbounded entity, especially when it is liquid,

and this pattern seems to become entrenched. This leads to the hypothesis

that recurrent use of coercion may change grammar eventually. I hope

that the increased acceptance of the coerced expressions in this experiment

supported this hypothesis.

From the perspective of the second langue acquisition, the result of

the current study calls for further study regarding the input quality and

frequency. Since English learners have not established their grammatical

knowledge and they may not be confident in their knowledge, the input

quality is very important (see Ellis et al. (2014) for the prototypicality

and salience of the input). Since coerced expressions are not perfectly

natural even for the NS, if NNS receive the coerced input excessively,

their grammar may gradually deviate from the grammar of NS. Note

that the current study aims to show that existing grammar can change

78 Soyeon Yoon

if speakers frequently experience the instances of the extended schemas

(i.e. coercion). For this reason, the experiment was designed in the way

that the participants experienced the coerced input unusually frequently.

This frequent coerced input did affect the grammar, as this study showed.

Then, in the second language acquisition, we need to consider the input

quality and frequency: Should the learners be given only “grammatical”

input or be given the input same as the NS (including coerced expressions

with similar frequency)? To study further about input quality and fre-

quency regarding coercion, I suggest considering the following aspects.

First, the ratio of coerced input to “grammatical” input must be considered:

when coerced input is more frequent than grammatical input (like the

current study where “grammatical” input was not given while coerced

input was given four times for each passage in the input session); when

coerced input is given with the same frequency pattern as found in the

corpus (e.g. give used in the DC is given 710 times while earn used in

the DC is given only once (see Yoon 2013)). In addition, the influence

of coerced input on NS should be examined because the flexibility towards

the extend schema may be different from the NNS.

6. Conclusion

The previous studies on the usage-based approaches to language acquis-

ition have shown that frequent experience of particular linguistic pattern

is important in forming grammar. Mostly, both in the first and second

language acquisition, the grammar in question was assumed to be the

one that the speakers newly acquire. In the current study, however, I

examined whether the existing grammar can change after the speakers

experience the input that does not match their linguistic knowledge.

Specifically, this study created sentences requiring various degrees of co-

ercion: In the sentences, the DC is used with the verbs of various degrees

of semantic compatibility. When the speakers read these sentences fre-

quently, they judged the sentences more acceptable. In other words, the

speakers extend the schemas of the verbs and the DC so that they can


use them together. Also, the speakers seemed to apply the extended sche-

mas to the sentences where the DC was used with other verbs which

were not used as input. The generalization that the speakers showed in

the experiment implies that the extended schema may be entrenched as

grammar if this pattern is repeated, and presents the possibility that speak-

ers change their existing linguistic knowledge on the basis of the input

that they experience.

The current experiment was conducted with the Korean participants,

but if it can be conducted with greater number of native speakers over

longer period of time, the results will further have a strong implication

in how language and grammar change through time.

References

Boyd, Jeremy K. and Adele E. Goldberg. (2009). Input effects within a con-

structionist framework. The Modern Language Journal 93, 418-429.

Croft, William. (1991). Syntactic Categories and Grammatical Relations: The Cognitive

Organization of Information. Chicago; London: The University of Chicago Press.

Ellis, Nick and Laura Collins. (2009). Input and second language acquisition: the

roles of frequency, form, and function. The Modern Language Journal 93,

329-335.

Ellis, Nick and Fernando Ferreira-Junior. (2009). Construction learning as a func-

tion of frequency, frequency distribution, and function. The Modern Language

Journal 93, 370-385.

Ellis, Nick C., Matthew B. O’Donnell, and Ute Römer. (2014). The processing of

verb-argument constructions is sensitive to form, function, frequency, con-

tingency, and prototypicality. Cognitive Linguistics 25, 55-98.

Goldberg, Adele E., Devin M. Casenhiser, and Nitya Sethuraman. (2004).

Learning argument structure generalizations. Cognitive Linguistics 15, 289-316.

Kidd, Evan, Elena V. M. Lieven, and Michael Tomasello. (2010). Lexical fre-

quency and exemplar-based learning effects in language acquisition: evidence

from sentential complements. Language Sciences 32, 132-142.

Langacker, Ronald. W. (1988). A usage-based model. In B. Rudzke-Ostyn ed.,

Topics in cognitive linguistics, 127-161. Amsterdam: Benjamins.

Langacker, Ronald. W. (1987). Foundations of Cognitive Grammar, Vol. 2. Stanford

80 Soyeon Yoon

University Press.

Levin, Beth. (1993). English Verb Classes and Alternations: A Preliminary Investigation.

Chicago, London: The University of Chicago Press.

Michaelis, Laura. A. (2005). Entity and event coercion in a symbolic theory of

syntax. In J. Östman, and M. Fried eds., Construction grammars: Cognitive

grounding and theoretical extensions, 45-88. Amsterdam; Philadelphia: John

Benjamins Publishing Company.

Pin͂ango, Maria. M., Aaron Winnick, Rashad Ullah, and Edgar Zurif. (2006).

Time-course of semantic composition: The case of aspectual coercion. Journal of

Psycholinguistic Research 35, 233-244.

Wierzbicka, Anna. (1996). Semantics: Primes and Universals. New York: Oxford

University Press.

Wonnacott, Elizabeth, Jeremy K. Boyd, Jennifer Thomson, and Adele E.

Goldberg. (2012). Input effects on the acquisition of a novel phrasal con-

struction in 5 years olds. Journal of Memory and Language 66, 458-478.

Year, Jungeun and Peter Gordon. (2009). Korean speakers’ acquisition of the

English ditransitive construction: the role of verb prototype, input distribution,

and frequency. The Modern Language Journal 93, 399-417.

Yoon, Soyeon. (2013). Correlation between semantic compatibility and frequency:

A usage-based approach. Linguistic Research 30.2, 243-272.

Yoon, Soyeon. (2012). Constructions, Semantic Compatibility, and Coercion: An

Empirical Usage-based Approach. Ph. D. Dissertation, Rice University.

Ziegeler, Debra. (2007). A word of caution on coercion. Journal of Pragmatics 39,

990-1028.

Soyeon Yoon

Department of English Language and Literature

Incheon National University

119 Academi-ro, 15-413

Yeonsu-gu, Incheon

Email: [email protected]

Received: February 2, 2016

Revised version received: April 8, 2016

Accepted: April 10, 2016


Appendix 1

Stimuli

(The sentences in Type “input” were used in the input sessions.)

Type Verb Sentence

Input give Jenny gave me a rose at the restaurant.

make Bill made her a toy for her birthday.

refuse Sam refused her a ring last night.

cut Jane cut him a belt on his birthday.

remember Beth remembered me a watch on their anniversary.

edible NP2 give John gave her a piece of cake this evening.

make Susan made him a cup of tea last weekend.

refuse Sam refused her a beer last night.

cut Ron cut her a slice of bread this morning.

remember Ron remembered her a meal this morning.

non-edible NP2 give John gave her a pencil this evening.

make Susan made him a bag last weekend.

refuse Beth refused him a knife last week.

cut Ron cut her a mirror on Friday.

remember Ron remembered her a shirt this morning.

Class1 lend Matt lent me a pencil yesterday.

(same class) cook Sam cooked them a meal last evening.

deny Julie denied me a drink last night.

slice Tom sliced me an onion in the evening.

imagine Sam imagined me a house three weeks ago.

Class2 send Julie sent him a letter last year.

(similar class) find Ben found her a table last night.

decline Beth declined him a seat this morning.

break John broke me a cracker after lunch.

stay Mark stayed her a chair on her birthday.

82 Soyeon Yoon

Appendix 2

The Sources of the Passages Used in the Input Sessions

< from Book Excerpts >Collins, M. (2008). Cidhina’s New Toy. Global Citizenship for Young Children.

Sage. (https://books.google.co.kr/books?id=9YHDFR1AhvoC&dq=Life+in+the

+countryside,+where+Cidhina+lives,&hl=ko&source=gbs_navlinks_s)

Hare, E. B. (2009). Skyscrapers: 365 Stories that Build You Up : Daily Devotions for

Juniors. Review and Herald Pub Assoc. (https://books.google.co.kr/books?id=

xPEVqNHBcgQC&pg=PA232&lpg=PA232&dq=One+day+when+my+eld-

est+son,+Lenny,+was+just+a+little+fellow+4+years+old&source=bl&ots=Aiqgr

SyWzn&sig=eMHRqCP6utW0N2h6zN4uzbxAJFQ&hl=ko&sa=X&ved=0ahUK

EwiW8pno0NXKAhXjxqYKHaQ3AKAQ6AEIGjAA#v=onepage&q=One%

20day%20when%20my%20eldest%20son%2C%20Lenny%2C%20was%20just%

20a%20little%20fellow%204%20years%20old&f=false)

Nee, W. (2011). Changed into his Likeness. CLC Publications. https://books.google.co.kr/

books?id=0MeQ6owTFBgC&pg=PT81&lpg=PT81&dq=You+can+answer+each

+question+with+absolute+assurance.&source=bl&ots=1CzRYYknPh&sig=bp4

ZSwetbo0x02p9z81m_-qtato&hl=ko&sa=X&ved=0ahUKEwi7oJPd0dXKAhXEIq

YKHQFaDT0Q6AEIIjAB#v=onepage&q=You%20can%20answer%20each%

20question%20with%20absolute%20assurance.&f=false)

Rajanna, N. (2010). The Diamond Ring. AuthorHouse. (https://books.google.co.kr/

books?id=l5peaQxvDbMC&pg=PA177&dq=if+this+letter+becomes+known+

to+your+husband.&hl=ko&sa=X&ved=0ahUKEwjBs-6g0tXKAhUCKqYKHS9O

DigQ6AEIGzAA#v=onepage&q=if%20this%20letter%20becomes%20known%

20to%20your%20husband.&f=false)

Lovring, M. (2013). The Remedy. A&C. Black. (https://books.google.co.kr/

books?id=7thXKF_Tn7oC&pg=PT207&dq=%22refused+me+a+ring%22&hl=ko

&sa=X&ved=0ahUKEwir_c6L09XKAhUFKKYKHT6jD8sQ6AEIGjAA#v=one-

page&q=%22refused%20me%20a%20ring%22&f=false)

Ferreri, J. (2011). For the Love of Wood / For the Love of Food. Author House.

(https://books.google.co.kr/books?id=f3x_FnU65UMC&pg=PA63&lpg=PA63&

dq=%22make+me+a+toy%22&source=bl&ots=nKDKV3UVT-&sig=RxmFp7wu

WDQNZLoG4GjR7AmMA2Q&hl=ko&sa=X&ei=WcxiVfGhH4K3mAWLy4KI

CQ&ved=0CF0Q6AEwCQ#v=onepage&q=%22make%20me%20a%20toy%22&f

=false)

Reid, M. J. and Sewid-Smith, D. (2007). Paddling to Where I Stand. UBC Press.


(https://books.google.co.kr/books?id=59NTCgAAQBAJ&pg=PA82&dq=This+

man+was+a+great+carver.&hl=ko&sa=X&ved=0ahUKEwjlqNTQztXKAhXkM

qYKHd63DVIQ6AEIUjAG#v=onepage&q=This%20man%20was%20a%

20great%20carver.&f=false)

Scott, B. (2011). In Love with the Enemy. Lulu.com. (https://books.google.co.kr/books?id=

MvNlAwAAQBAJ&pg=PA65&dq=when+given+to+a+lover+a+four-leaf+clover+

stood+for+good+luck+and+be+mine.&hl=ko&sa=X&ved=0ahUKEwjm87nvztX

KAhUj2qYKHS5bDtkQ6AEIHTAA#v=onepage&q=when%20giv-

en%20to%20a%20lover%20a%20four-leaf%20clover%20stood%20for%20good%

20luck%20and%20be%20mine.&f=false)

Lipthay, M. W. (2012). Eldorado: My Childhood During the Great Depression.

AuthorHouse. (https://books.google.co.kr/books?id=ZNbcG-5B0uQC&pg=PA55&dq=

nothing+fancy,+just+a+tiny+sailboat+that+could+be+sailed+or+pulled&hl=ko&sa=

X&ved=0ahUKEwiU_MaLz9XKAhXi5aYKHVcRDxgQ6AEIGzAA#v=onepage&q=

nothing%20fancy%2C%20just%20a%20tiny%20sailboat%20that%20could%20be%

20sailed%20or%20pulled&f=false)

<from Blogs and Forums>DIY Fixes for Shoes, Purses and Belts. May 01, 2009. (https://www.mainstreet.com/

article/diy-fixes-shoes-purses-and-belts)

http://glassmagazine.com/glassblog/more-americas-funniest-glass-stories-117727

http://forum.m1911.org/archive/index.php/t-53414.html

http://southmagazine.co.za/articles/arts-culture/toys-from-tyres/

<from Corpus of Contemporary American English>COCA_FIC_ Legerdemain

COCA_MAG_The Untold Jackie

84 Soyeon Yoon

Appendix 3

Descriptive Statistics of the Preliminary Test on Native Speakers

(N = 7) and Pretest on Non-native Speakers (N = 58)

Type Verb Speakers Mean Standard deviation

Input

give NS 1.34 0.15

　 NNS 1.95 0.79

make NS 2.09 1.57

　 NNS 2.19 1.73

refuse NS 5.66 1.07

　 NNS 4.12 1.84

cut NS 5.60 1.57

　 NNS 5.03 1.50

remember NS 7.00 0.15

　 NNS 4.50 1.56

edible NP2

give NS 2.09 2.15

　 NNS 1.48 1.23

make NS 2.14 0.15

　 NNS 2.03 1.45

refuse NS 4.80 2.21

　 NNS 4.76 1.65

cut NS 4.71 1.04

　 NNS 4.41 2.10


　 NNS 4.60 1.46

non-edible NP2

give NS 2.09 1.07

　 NNS 1.48 1.33

make NS 2.14 1.63

　 NNS 2.03 2.06

refuse NS 4.80 1.78

　 NNS 4.76 1.26

cut NS 4.71 1.75

　 NNS 4.41 1.30


　 NNS 4.60 1.67


Type Verb Speakers Mean Standard deviation

Class1 (same class)

lend NS 1.54 0.15

NNS 1.60 1.22

cook NS 3.06 2.19

NNS 4.64 1.82

deny NS 4.11 1.92

NNS 4.22 1.61

slice NS 6.06 1.38

NNS 4.03 1.92

imagine NS 7.00 2.35

NNS 5.03 1.52

Class2 (similar class)

send NS 1.34 0.57

NNS 1.78 1.01

find NS 3.46 2.42

NNS 2.88 1.71

decline NS 3.17 1.92

NNS 4.31 1.67

break NS 2.11 1.04

NNS 3.31 1.74

stay NS 5.40 0.00

NNS 5.05 1.50

Influence of Coerced Input on English Learners’ Grammar

Documents

Transcript of Influence of Coerced Input on English Learners’ Grammar